Inflight Documentation
Documentation/Sample Report

Simulation Guidance Report

See how Inflight delivers actionable optimization recommendations with full transparency. Every suggestion includes evidence, predictions, and safety analysis.

WARNINGValidate in Staging

Recommendation ready with minor risk factors

78%
Confidence
83%
Win Probability
Service
payment-api
Runtime
Java 21 (OpenJDK)
Fidelity Mode
FULL (DES)
Generated
2 minutes ago

What You Get

Inflight's Simulation Guidance Reports provide everything you need to make informed optimization decisions. No black boxes—just clear, evidence-based recommendations.

51
Candidates Evaluated
Comprehensive search across the optimization space
78%
Confidence Score
Statistical certainty in the recommendation
83%
Win Probability
Likelihood of outperforming baseline
3
Top Candidates
Pareto-optimal configurations identified

Report Structure

Every report follows a consistent structure designed for quick decision-making while providing deep-dive details when needed.

Decision Summary

Clear verdict with confidence score and win probability against baseline configuration.

Multi-Candidate Comparison

Side-by-side analysis of top candidates from the Pareto frontier with trade-off visibility.

Metric Predictions

Detailed before/after predictions for CPU, memory, throughput, GC, and response times.

Safety Analysis

Specific violations flagged with severity levels and mitigation recommendations.

Recommended Patch

Ready-to-apply Kubernetes YAML or runtime configuration changes.

Monitoring Plan

Post-deployment thresholds and metrics to watch for validation.

Multi-Candidate Comparison

Inflight evaluates 68 candidates and surfaces the top performers from the Pareto frontier—configurations that represent optimal trade-offs between competing objectives like latency, throughput, and resource usage.

Configuration
P99 Latency
Throughput
GC Pause
Memory Usage
BaselineCurrent
2048m heap, G1GC
245ms
1,250 req/s
85ms
82%
Candidate A
1536m heap, G1GC
198ms
-19%
1,180 req/s
-6%
42ms
-51%
72%
-12%
Candidate BRecommended
1792m heap, ZGC
178ms
-27%
1,220 req/s
-2%
12ms
-86%
75%
-9%
Candidate C
1280m heap, G1GC
268ms
+9%
1,150 req/s
-8%
38ms
-55%
68%
-17%
Why Candidate B?

Candidate B offers the best balance: 27% latency improvement with only 2% throughput trade-off. The switch to ZGC dramatically reduces GC pause times by 86%, which is critical for payment processing where consistent response times matter more than raw throughput.

Detailed Metric Predictions

Comprehensive before/after analysis across all performance dimensions, with confidence scores for each prediction.

Latency

P50 Response Time
45ms
38ms
-16%
92%
P95 Response Time
180ms
142ms
-21%
89%
P99 Response Time
245ms
178ms
-27%
85%
P99.9 Response Time
520ms
395ms
-24%
78%

Throughput

Requests/Second
1,250
1,220
-2%
91%
Peak Throughput
1,890
1,850
-2%
88%
Error Rate
0.12%
0.09%
-25%
94%

Memory

Heap Usage (Avg)
1.72GB
1.34GB
-22%
93%
Heap Usage (Peak)
2.01GB
1.68GB
-16%
90%
Metaspace
128MB
128MB
0%
98%
Memory Headroom
12%
18%
+50%
87%

Garbage Collection

GC Pause (Avg)
85ms
12ms
-86%
91%
GC Pause (Max)
245ms
28ms
-89%
88%
GC Frequency
4.2/min
8.1/min
+93%
94%
GC CPU Overhead
8.2%
3.1%
-62%
89%

CPU

CPU Usage (Avg)
45%
48%
+7%
95%
CPU Usage (Peak)
78%
82%
+5%
92%
CPU Throttling
2.1%
1.8%
-14%
88%
14
Metrics Improved
2
Metrics Unchanged
1
Metrics to Monitor

Safety Analysis

Every recommendation is validated against your service's constraints. Violations are flagged with specific codes, descriptions, and actionable recommendations.

WARNINGINSUFFICIENT_MEMORY_HEADROOM

Predicted memory headroom of 18% is below the recommended 20% buffer for production workloads.

Threshold
Minimum 20% headroom required
Actual
18% predicted
Recommendation
Consider increasing memory limit by 200Mi or monitoring closely during peak traffic periods.
INFOTHROUGHPUT_REGRESSION_MINOR

Minor throughput reduction of 2% detected. Within acceptable tolerance for latency improvements.

Threshold
Maximum 5% reduction allowed
Actual
2% reduction predicted
Recommendation
Trade-off is acceptable given 27% P99 latency improvement. Monitor during rollout.

Safety Verdicts

Every recommendation receives a clear safety verdict based on comprehensive simulation analysis against your service's constraints.

APPROVED

All safety thresholds met. Safe to deploy with confidence.

  • All metrics within bounds
  • No critical violations
  • High confidence predictions
WARNING

Some risk factors detected. Validate in staging environment first.

  • Minor threshold violations
  • Trade-offs identified
  • Monitoring recommended
REJECT

Critical issues detected. Change will not achieve intended outcome.

  • Critical violations found
  • Unacceptable regression
  • Insufficient data quality

Ready-to-Apply Changes

Reports include the exact configuration changes needed—whether that's a Kubernetes patch, JVM flags, or Go environment variables. Copy, review, and apply.

JVM Configuration
  • GC Algorithm:G1GC → ZGC (Generational)
  • Heap Size:2048m → 1792m (-12%)
  • Metaspace:256m (unchanged)
Kubernetes Resources
  • CPU Request:500m → 600m (+20%)
  • CPU Limit:1000m → 1200m (+20%)
  • Memory Limit:2.5Gi → 2.4Gi (-4%)

Post-Deployment Monitoring

Every report includes a monitoring plan with specific thresholds to watch after deployment, ensuring you catch any unexpected behavior early.

Alert Thresholds

  • P99 Latency:Alert if > 220ms for 5 min
  • Memory Usage:Alert if > 85% of limit
  • Throughput:Alert if < 1,100 req/s
  • Error Rate:Alert if > 0.5%

Validation Period

  • First 4 hours:Active monitoring, quick rollback ready
  • 24 hours:Observe full traffic cycle
  • 7 days:Full validation, model calibration

Rollback Guidance

If any threshold is breached, revert to the baseline configuration immediately. Every report includes the original configuration preserved in the evidence appendix, along with step-by-step rollback instructions specific to your deployment method.

Evidence Appendix

Every recommendation is backed by traceable evidence. The appendix provides links to the specific data points, model parameters, and simulation results.

Historical Metrics

  • 7-day P99 latency trend analysis
  • Memory allocation patterns during peak hours
  • GC pause distribution histogram
  • Throughput correlation with heap size

Model Calibration

  • Last calibration: 2 hours ago
  • Training data: 14 days of production metrics
  • Model accuracy: 94.2% (backtested)
  • Drift detection: No significant drift

Simulation Parameters

  • Fidelity mode: FULL (Discrete Event Simulation)
  • Simulation duration: 24-hour equivalent
  • Traffic pattern: Production replay
  • Confidence intervals: 95%

Full Traceability

Every prediction links back to the historical metrics, model calibration data, and simulation parameters that generated it.

Confidence Intervals

Predictions include 95% confidence intervals so you understand the range of expected outcomes, not just point estimates.

See It In Action

Ready to get Simulation Guidance Reports for your own services? Connect your APM and start receiving intelligent optimization recommendations.

Related Documentation